Various network configurations may involve multiple computers sharing a single network link. In an illustrative configuration, an enterprise may deploy centralized servers (e.g., in a data center) for hosting corporate data, applications, and services. One or more remote users (e.g., at a branch office) may then access the data center over a single network link, such as a wide area network (“WAN”) link. For example, a server computer in Austin, Tex. may host a video conference call accessed by multiple users in an Atlanta, Ga. branch office. In this case, the users in Atlanta may share a WAN link within the branch office.
The effectiveness of the network link may depend on the amount of bandwidth utilized on the network link. In particular, overconsumption of bandwidth may result in congestion at the network link, potentially causing the network link to lose data packets during transmission. One approach to resolving congestion on the network link is to transmit redundant data in order to compensate for lost data packets. For example, forward error correction (“FEC”) or another suitable mechanism may be utilized. However, FEC may cause additional congestion on the network by introducing additional data traffic onto the network link. Further, transmitting redundant data does not address the problem of congestion.
Another approach to resolving congestion on the network link is to dynamically manage the amount of data traffic transmitted on the network link. In particular, if congestion is detected on the network link, then the amount of data traffic transmitted across the network may be reduced in an attempt to reduce the congestion. However, conventional techniques for determining the presence of congestion on the network link have drawbacks. For example, one way to determine the presence of congestion on the network link involves utilizing a conventional link speed detection process. A conventional link speed detection process measures link capacity across the network link and is generally adequate for detecting a bottleneck in the last hop (e.g., a router) on the network link. However, the link speed detection process may fail to detect the presence of congestion across the entire network link.
It is with respect to these considerations and others that the disclosure made herein is presented.